Posted by Steven Vancoillie on April 01, 2015 at 09:17:31:

In Reply to: Re: CASPT2 convergence posted by Roser on March 09, 2015 at 15:42:27:

Hi,

: I have a new question: What does the CASPT2 iteration do?

: A perturbation does not need an iterative procedure. I have read in the manual: "Of course, the complete Hˆ0 is used to solve the CASPT2 equations, which is why an iterative procedure is needed. "

: What is the program doing?

I think you are confusing iterative procedure with variational, in the sense that perturbation theory does not have the true energy as a lower bound.

The program is iteratively solving an equation system Ax=b, where A is the zeroth-order hamiltonian, x are the unknown parameters (that define the first-order wavefunction) and b is the right-hand-side expression. The method is called "conjugate gradient". You can look in the CASPT2 paper(s) for more information.

: And you asked for the output (sorry for the delay). I copy the state without convergence (not all iterations). If you need something more to understand the out just tell me.

The very low reference weight might indicate one or more intruder states. After that, caspt2 should print an analysis of coefficients/denominators that give you a clue if that could be the case. Can you provide those? You can also email me the entire (caspt2) output of one of your calculations.

How did you choose the Dy/Er active space, which states are you computing, etc. ?